On 9/21/2026 7:29 PM, Lazar, Lijo wrote:


On 21-Sep-26 7:15 PM, James Zhu wrote:
amdgpu_userq_input_trap_params_validate() ends up in
amdgpu_userq_input_va_validate(), which walks the VM mapping tree and
updates bo_va->userq_va_mapped, so it requires the caller to hold the
vm->root.bo reservation. The CWSR branch of mes_userq_mqd_create() called it without that reservation, tripping the dma_resv_assert_held() on every
compute queue creation when trap handling is enabled:

  WARNING: drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c:232 at
  amdgpu_userq_input_va_validate+0xb9/0xd0 [amdgpu]
  Call Trace:
   mes_userq_mqd_create+0x49c/0x9e0 [amdgpu]
   amdgpu_userq_ioctl+0x935/0x1070 [amdgpu]
   drm_ioctl_kernel+0xb3/0x110 [drm]

Bracket the call with amdgpu_bo_reserve()/amdgpu_bo_unreserve() like the
other input VA validations in the same function.

Fixes: 7ca9abf4ac38 ("drm/amdgpu: Add user save area params to mqd input")


Hi Alex/Srini,

Which branch are these patches based on? I don't see any cwsr patch upstream yet, probably that makes Fixes irrelevant.

These patches are based on |<exact-branch-name>| from the unified-interface tree, not upstream drm-next. The CWSR/userq changes are already present in that branch.

Thanks, Srini

Reply via email to